[PATCH] vfio: pci: fix oops in case of vfio_msi_set_vector_signal failure

[Alex Williamson]

On Mon, 2016-02-01 at 18:27 +0100, Eric Auger wrote:
> Hi Alex,
> On 01/29/2016 10:41 PM, Alex Williamson wrote:
> > On Fri, 2016-01-29 at 14:43 +0000, Eric Auger wrote:
> > > In case vfio_msi_set_vector_signal fails we tear down everything.
> > > In the tear down loop we compare int j against unsigned start. Given
> > > the arithmetic conversion I think it is converted into an unsigned and
> > > becomes 0xffffffff, leading to the loop being entered again and things
> > > turn bad when accessing vdev->msix[vector].vector. So let's use int
> > > parameters instead.
> > >  
> > > Signed-off-by: Eric Auger <eric.auger at linaro.org>
> > > ---
> > >  drivers/vfio/pci/vfio_pci_intrs.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > >  
> > > diff --git a/drivers/vfio/pci/vfio_pci_intrs.c b/drivers/vfio/pci/vfio_pci_intrs.c
> > > index 3b3ba15..510c48d 100644
> > > --- a/drivers/vfio/pci/vfio_pci_intrs.c
> > > +++ b/drivers/vfio/pci/vfio_pci_intrs.c
> > > @@ -374,8 +374,8 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_device *vdev,
> > >  	return 0;
> > >  }
> > >  
> > > -static int vfio_msi_set_block(struct vfio_pci_device *vdev, unsigned start,
> > > -			      unsigned count, int32_t *fds, bool msix)
> > > +static int vfio_msi_set_block(struct vfio_pci_device *vdev, int start,
> > > +			      int count, int32_t *fds, bool msix)
> > >  {
> > >  	int i, j, ret = 0;
> > >  
> > 
> > Nice find, I don't think that's the only bug there though.  If @start is
> > -1 (UINT32_MAX) and @count is 1, then @j gets set to -1 in the setup and
> > we hit the same index dereference problem.  What if we did this instead:
> > 
> > diff --git a/drivers/vfio/pci/vfio_pci_intrs.c b/drivers/vfio/pci/vfio_pci_intrs.c
> > index 3b3ba15..2ae84ad 100644
> > --- a/drivers/vfio/pci/vfio_pci_intrs.c
> > +++ b/drivers/vfio/pci/vfio_pci_intrs.c
> > @@ -309,14 +309,14 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_device *vdev,
> >  				      int vector, int fd, bool msix)
> >  {
> >  	struct pci_dev *pdev = vdev->pdev;
> > -	int irq = msix ? vdev->msix[vector].vector : pdev->irq + vector;
> > -	char *name = msix ? "vfio-msix" : "vfio-msi";
> >  	struct eventfd_ctx *trigger;
> > -	int ret;
> > +	int irq, ret;
> >  
> > -	if (vector >= vdev->num_ctx)
> > +	if (vector < 0 || vector >= vdev->num_ctx)
> >  		return -EINVAL;
> >  
> > +	irq = msix ? vdev->msix[vector].vector : pdev->irq + vector;
> > +
> >  	if (vdev->ctx[vector].trigger) {
> >  		free_irq(irq, vdev->ctx[vector].trigger);
> >  		irq_bypass_unregister_producer(&vdev->ctx[vector].producer);
> > @@ -328,8 +328,9 @@ static int vfio_msi_set_vector_signal(struct vfio_pci_device *vdev,
> >  	if (fd < 0)
> >  		return 0;
> >  
> > -	vdev->ctx[vector].name = kasprintf(GFP_KERNEL, "%s[%d](%s)",
> > -					   name, vector, pci_name(pdev));
> > +	vdev->ctx[vector].name = kasprintf(GFP_KERNEL, "vfio-msi%s[%d](%s)",
> > +					   msix ? "x" : "", vector,
> > +					   pci_name(pdev));
> >  	if (!vdev->ctx[vector].name)
> >  		return -ENOMEM;
> >  
> > @@ -379,7 +380,7 @@ static int vfio_msi_set_block(struct vfio_pci_device *vdev, unsigned start,
> >  {
> >  	int i, j, ret = 0;
> >  
> > -	if (start + count > vdev->num_ctx)
> > +	if (start >= vdev->num_ctx || start + count > vdev->num_ctx)
> >  		return -EINVAL;
> >  
> >  	for (i = 0, j = start; i < count && !ret; i++, j++) {
> > @@ -388,7 +389,7 @@ static int vfio_msi_set_block(struct vfio_pci_device *vdev, unsigned start,
> >  	}
> >  
> >  	if (ret) {
> > -		for (--j; j >= start; j--)
> > +		for (--j; j >= 0 && j >= start; j--)
> >  			vfio_msi_set_vector_signal(vdev, j, -1, msix);
> >  	}
> >  
> > 
> > So we fix the problem with vfio_msi_set_vector_signal() dereferencing
> > the array before it validates the index (even though it shouldn't be
> > able to get there anymore), and then we do a better job of verifying
> > start and count (comparing to num_ctx will use unsigned even though
> > num_ctx itself is signed) and finally explicitly test the <0 case, which
> > I suppose we could also do by casting start at that point (we know it's
> > within the bounds of a signed integer given the previous tests).
> 
> Yes it looks OK to me.
> 
> I guess you submit? I will test it.

Yep, I'll post a real patch.  Thanks,

Alex